Identification of potential key genes in gastric cancer using bioinformatics analysis.

Gastric cancer (GC) is one of the most common types of cancer worldwide. Patients must be identified at an early stage of tumor progression for treatment to be effective. The aim of the present study was to identify potential biomarkers with diagnostic value in patients with GC. To examine potential therapeutic targets for GC, four Gene Expression Omnibus (GEO) datasets were downloaded and screened for differentially expressed genes (DEGs). Gene Ontology and Kyoto Encyclopedia of Genes and Genomes (KEGG) analyses were subsequently performed to study the function and pathway enrichment of the identified DEGs. A protein-protein interaction (PPI) network was constructed. The CytoHubba plugin of Cytoscape was used to calculate the degree of connectivity of proteins in the PPI network, and the two genes with the highest degree of connectivity were selected for further analysis. Additionally, the two DEGs with the largest and smallest log Fold Change values were selected. These six key genes were further examined using Oncomine and the Kaplan-Meier plotter platform. A total of 99 upregulated and 172 downregulated genes common to all four GEO datasets were screened. The DEGs were primarily enriched in the Biological Process terms: 'extracellular matrix organization', 'collagen catabolic process' and 'cell adhesion'. These three KEGG pathways were significantly enriched in the categories: 'ECM-receptor interaction', 'protein digestion and absorption', and 'focal adhesion'. Based on Oncomine, expression of ATP4A and ATP4B were downregulated in GC, whereas expression of the other genes were all upregulated. The Kaplan-Meier plotter platform confirmed that upregulated expression of the identified key genes was significantly associated with worse overall survival of patients with GC. The results of the present study suggest that FN1, COL1A1, INHBA and CST1 may be potential biomarkers and therapeutic targets for GC. Additional studies are required to explore the potential value of ATP4A and ATP4B in the treatment of GC. Copyright: © Wang et al.

Introduction

Gastric cancer (GC) is a malignant tumor that originates in the epithelium of the gastric mucosa and is one of the most common types of malignant tumors in the world (1). According to GLOBOCAN 2018, there were >1,000,000 new cases of GC and ~783,000 deaths in 2018, thus making it the cancer type with the fifth highest incidence rate and the third highest mortality in the world (2). The poor five-year survival rate of GC is primarily due the advanced stage of gastric tumors at the initial diagnosis in the majority of patients, and thus limits treatment opportunities (3). According to the Cancer Staging Manual, 8th edition, of the American Joint Committee on Cancer, only 30% of GC cases are diagnosed prior to metastasis, and the five-year survival for pathological Tumor-Node-Metastasis stage groups are between 68-80% for stage I, 46-60% for stage II, 8-30% for stage III and 5% for stage IV (4). Therefore, identifying potential biomarkers for patients with early GC is critical for improving patient outcomes.

In recent years, a variety of bioinformatics methods have contributed greatly to the discovery of biomarkers associated with tumor development, diagnosis and prognosis (5-8). The combined use of multiple databases of biological information for the analysis of cancer has also yielded certain breakthroughs. Yong et al (9) used Gene Expression Omnibus (GEO), Oncomine, Search Tool for Recurring Instances of Neighbouring Genes (STRING) and other databases for bioinformatic analysis, and concluded that PPP2CA may function as an oncogene and a prognostic biomarker or therapeutic target in the progression of colorectal cancer. Troiano et al (10) used the GEO database and Oncomine to examine the expression of BIRC5/Survivin in oral squamous cell carcinoma and showed that Survivin expression was upregulated compared with non-cancerous tissue. In addition, immunohistochemistry staining showed that cytoplasmic expression of Survivin was associated with poor overall survival in patients with oral squamous cell carcinoma. It may be beneficial to use multiple datasets and analysis tools to determine the potential mechanisms underlying development and progression of GC, and to identify potentially novel and specific diagnostic biomarkers for early detection of GC to improve the survival of patients.

In the present study, the expression profiles from four datasets (GSE13911, GSE19826, GSE54129 and GSE118916) in human GC and normal gastric tissue samples were obtained from the GEO database and analyzed to identify differentially expressed genes (DEGs). Gene Ontology (GO) and pathway enrichment analysis were performed to identify the biological functions and pathways of the DEGs. STRING and Cytoscape were used to construct a protein-protein interaction (PPI) network, and a total of six key genes were selected from the PPI network and DEGs. The value of the key genes was validated using the Oncomine and Kaplan-Meier platforms to further increase the reliability of the results and confirm the prognostic value of the key genes.

Materials and methods

Microarray data

The key word ‘gastric cancer’ was searched in the GEO database (ncbi.nlm.nih.gov/geo/), and a total of 9,224 datasets on human GC were retrieved. In the present study, four gene expression profiles from the GEO database were used, as they have not been studied together previously. The four datasets were: GSE13911(11), GSE19826(12), GSE54129 and GSE118916(13). Among these, GSE13911, GSE19826 and GSE54129 were based on the GPL570 platform [(HG-U133_Plus_2) Affymetrix Human Genome U133 Plus 2.0 Array]. GSE118916 was based on the GPL15207 platform [(PrimeView) Affymetrix Human Gene Expression Array].

Identification of DEGs

DEGs between GC samples and normal controls were identified using the GEO2R online analysis tool (ncbi.nlm.nih.gov/geo/geo2r); |log FC|≥1.0 and corrected P<0.05 were used as the cutoff criteria. The common DEGs of the four gene expression profiles were screened using Wayne analysis in Funrich (funrich.org/).

GO and KEGG enrichment analyses of DEGs

After obtaining the common DEGs, GO (14,15) and KEGG (16) analyses of the DEGs were performed using the Database for Annotation Visualization and Integrated Discovery (DAVID) online tool (17,18), with P<0.01 used as the threshold for significance. GO was used to identify the enrichment functions of three independent categories of genes; biological process (BP), cellular component (CC) and molecular function (MF). KEGG was used to search for the pathways associated with the identified genes (19). Only the top 10 BP, CC and MF terms, and the KEGG pathway with the smallest P-value were selected for further examination in the present study. The figures were generated using the OmicShare tools (omicshare.com/tools), a free online platform for data analysis.

PPI network construction

To explore the interaction between DEGs, the DEGs were analyzed using STRING (20) to generate a PPI network. PPI pairs with a combined score >0.4 were extracted, and disconnected nodes in the network were hidden. Subsequently, the PPI network was visualized using Cytoscape (21) and the degree of each protein node was calculated using the cytoHubba (22) plug-in in Cytoscape.

Identification of key genes

The two genes with the highest degree of connectivity in the PPI network, the two genes with the largest logFC values and the two genes with the smallest logFC among the shared DEGs were selected and considered key genes.

Analysis of key genes in Oncomine

The Oncomine database (oncomine.org/) was used to explore the mRNA expression differences of six key genes between GC and normal gastric tissue. Oncomine is a chip-based gene database and integrated data mining online cancer microarray database designed to facilitate the discovery of novel biomarkers from genome-wide expression analysis (23).

Survival analysis of key genes

The Kaplan-Meier plotter (24) is an online tool that can assess the effect of 54,000 genes on survival in 21 types of cancer. The largest datasets include breast (n=6,234), ovarian (n=2,190), lung (n=3,452) and gastric cancer (n=1,440) cancer. The primary purpose of the tool is to discover and validate biomarkers for survival. Online survival analysis of the selected key genes based on the GC database was performed using Kaplan-Meier Plotter. The hazard ratio (HR) with 95% confidence intervals (CIs) and log-rank P-values were calculated.

Results

GSE13911 includes 38 GC samples and 31 normal samples, GSE19826 contains 12 GC samples and 15 normal samples, GSE54129 contains 111 GC samples and 21 normal samples, and GSE118916 contains 15 GC samples and 15 normal samples (Table I). In GSE13911, there are 26 intestinal, 4 mixed, 6 diffuse and 2 unclassified gastric carcinoma tissues, as well as 31 normal adjacent tissues. Unfortunately, information on the histological subtypes were not available in the other datasets. In the datasets, 1,001 upregulated and 2,304 downregulated DEGs were identified in GSE13911, 407 upregulated and 753 downregulated DEGs were identified in GSE19826, 1,852 upregulated and 2,083 downregulated DEGs were identified in GSE54129, and 977 upregulated and 903 downregulated DEGs were identified in GSE118916. Wayne analysis identified 99 common upregulated genes and 172 common downregulated genes were obtained from the 4 datasets (Table II; Fig. 1).

Table I

Information for four gene expression profiles from Gene Expression Omnibus.

Dataset ID	Gastric cancer	Normal	Total Number	Platform
GSE13911	38	31	69	GPL570
GSE19826	12	15	27	GPL570
GSE54129	111	21	132	GPL570
GSE118916	15	15	30	GPL15207

Table II

The differentially expressed genes identified from the four gene expression profiles, between gastric cancer and normal tissues.

Differentially expressed genes	Gene terms
Upregulated	INHBA CST1 COL11A1 FAP COL10A1 FNDC1 COL8A1 SERPINH1 CDH3 THBS2 CLDN1 TNFRSF11B SPP1 COL1A2 SFRP4 SULF1 CPXM1 BMP1 MFAP2 COL1A1 CTHRC1 BGN RARRES1 IGF2BP3 THBS4 COL6A3 SRPX2 OSR2 HOXB7 TIMP1 ASPN THY1 FKBP10 PRRX1 SDS APOE PMEPA1 COL12A1 GPNMB FBN1 ADAM12 C3 APOC1 COL5A1 SPARC EPHB2 NID2 CMTM3 PLEKHO1 TNFRSF10B EHD2 FN1 MMP11 COCH AMIGO2 COL5A2 OLFML2B KLHL23 SPOCK1 CDH11 TWIST1 RAB31 SULF2 FGD6 VCAN ITGBL1 PCOLCE HAVCR2 THBS1 DNM1 IGFBP7 PLAU TMEM158 COL3A1 FLNA EDNRA LEF1 LIPG FZD2 GXYLT2 S100A10 LGALS1 NRP2 SIRPA ANTXR1 CD9 LIF COL4A2 TGM2 COL6A1 PDPN KCNJ8 ACTN1 GPR161 ZAK RCN3 BAG2 BHLHE40 COL4A1
Downregulated	ATP4A ATP4B KCNE2 AQP4 GIF LIPF GKN1 GKN2 DPCR1 PGC SOSTDC1 ESRRG MUC6 SST FBP2 CPA2 VSIG1 CXCL17 PDIA2 CCKBR TMED6 CHGA TFF2 PSCA FUT9 CA9 SCNN1G GUCA2B C16orf89 SLC26A9 KLK11 CWH43 DNER PSAPL1 CNTN3 ALDH3A1 GATA5 SCGB2A1 UGT2B15 RDH12 CLIC6 NRG4 CLDN18 CAPN9 SLC16A7 SSTR1 FBXL13 TCN1 VSIG2 AKR1B10 B3GNT6 FOLR1 MUM1L1 CHGB MAL TRIM50 AKR7A3 KIAA1324 PAIP2B SULT2A1 PTPRZ1 ARX LIFR ALDH1A1 HYAL1 BEX5 CA2 CYP2C18 ME1 SCNN1B ADH7 GCNT2 ACER2 FMO5 HPGD RASSF6 TFF1 TMEM171 CA4 KCNJ16 LDHD KCNJ15 GABRB3 HOMER2 TMPRSS2 LYPD6B KLHDC7A ARHGAP42 PLAC8 IGFBP2 CAPN13 SYTL5 PDGFD RNASE1 RORC CYP2C9 EPN3 PBLD METTL7A ZBTB7C UBL3 SH3RF2 RNASE4 ARHGEF37 ALDH6A1 RAB27B SULT1B1 PKIB PXMP2 GPRC5C RIMBP2 ATP8A1 FAM20A PIGR GOLM1 CYP3A5 FAM46C C9orf152 COBLL1 FA2H SORBS2 DGKD SGK2 TMEM220 ANG PLLP MYCN C1orf116 FGD4 SLC41A2 ADAM28 MAGI1 GRAMD1C IQGAP2 GULP1 SYTL2 DHRS7 OASL RNF128 DBT ELL2 RAB27A NOSTRIN NEDD4L PPFIBP2 AKR1C3 PELI2 SMPD3 PTPRN2 RASEF TMEM92 ABCC5 GALNT12 LMO4 NTN4 TMEM116 ID4 ELOVL6 ALDOB EPB41L4B CD36 GALNT5 SH3BGRL2 MAGI3 MICALL1 HIPK2 MAOA WWC1 SLC7A8 CDC14B FAM107B SUCLG2

Upregulated genes are listed from largest to smallest fold change values. Downregulated genes are listed from smallest to largest fold change values.

Figure 1.

Venn diagram of shared differentially expressed genes. (A) Upregulated and (B) downregulated genes from four gene expression profiles.

GO and KEGG pathway enrichment analyses of DEGs

GO and KEGG pathway enrichment analyses of the DEGs was performed using the online tool DAVID, and the results are presented in Table III. GO analysis showed that in BP, the DEGs were primarily enriched for the GO terms: ‘extracellular matrix organization’, ‘collagen catabolic process’, ‘cell adhesion’, ‘collagen fibril organization’ and ‘digestion’ (Table III; Fig. 2A). CC analysis revealed that the DEGs were significantly enriched for the terms: ‘extracellular space’, ‘extracellular matrix’, ‘extracellular exosome’, ‘extracellular region’ and ‘endoplasmic reticulum lumen’ (Table III; Fig. 2B). For MF, the DEGs were enriched for the GO terms: ‘platelet-derived growth factor binding’, ‘collagen binding’, ‘extracellular matrix binding’, ‘inward rectifier potassium channel activity’ and ‘SMAD binding’ (Table III; Fig. 2C). According to KEGG pathway analysis, the DEGs were primarily enriched for the pathway terms: ‘ECM-receptor interaction’, ‘protein digestion and absorption’, ‘focal adhesion’, ‘amoebiasis’ and ‘gastric acid secretion’ (Table III; Fig. 2D).

Table III

GO term and KEGG pathway enrichment analyses of the 271 differentially expressed genes.

Category	Term	Description	Count	P-Value
BP term	GO:0030198	Extracellular matrix organization	23	1.28x10-13
BP term	GO:0030574	Collagen catabolic process	14	7.06x10-12
BP term	GO:0007155	cell adhesion	30	3.59x10-11
BP term	GO:0030199	Collagen fibril organization	9	7.87x10-08
BP term	GO:0007586	Digestion	10	3.19x10-07
BP term	GO:0035987	Endodermal cell differentiation	7	2.13x10-06
BP term	GO:0001501	Skeletal system development	11	3.42x10-05
BP term	GO:0008202	Steroid metabolic process	7	3.60x10-05
BP term	GO:0071230	Cellular response to amino acid stimulus	7	6.04x10-05
BP term	GO:0006805	Xenobiotic metabolic process	8	1.45x10-04
BP term	GO:0042060	Wound healing	8	1.70x10-04
BP term	GO:0006081	Cellular aldehyde metabolic process	4	4.70x10-04
BP term	GO:0030277	Maintenance of gastrointestinal epithelium	4	6.20x10-04
BP term	GO:0010107	Potassium ion import	5	6.98x10-04
BP term	GO:0007584	Response to nutrient	7	7.50x10-04
BP term	GO:0002576	Platelet degranulation	8	7.99x10-04
BP term	GO:0060021	Palate development	7	8.64x10-04
BP term	GO:0010812	Negative regulation of cell-substrate adhesion	4	0.001003
BP term	GO:0001503	Ossification	7	0.001131
BP term	GO:0030168	Platelet activation	8	0.001523
BP term	GO:0051216	Cartilage development	6	0.001703
BP term	GO:0010628	Positive regulation of gene expression	12	0.001721
BP term	GO:0001523	Retinoid metabolic process	6	0.001977
BP term	GO:0016525	Negative regulation of angiogenesis	6	0.002125
BP term	GO:0055114	Oxidation-reduction process	19	0.002857
BP term	GO:0032964	Collagen biosynthetic process	3	0.003084
BP term	GO:0008284	Positive regulation of cell proliferation	16	0.003752
BP term	GO:0001649	Osteoblast differentiation	7	0.004274
BP term	GO:0022617	Extracellular matrix disassembly	6	0.005144
BP term	GO:0071711	Basement membrane organization	3	0.005647
BP term	GO:0050891	Multicellular organismal water homeostasis	3	0.005647
BP term	GO:0001525	Angiogenesis	10	0.005716
BP term	GO:0042476	Odontogenesis	4	0.007007
BP term	GO:0010575	Positive regulation of vascular endothelial growth factor production	4	0.007007
BP term	GO:0050909	Sensory perception of taste	4	0.008568
BP term	GO:0001937	Negative regulation of endothelial cell proliferation	4	0.008568
BP term	GO:0040037	Negative regulation of fibroblast growth factor receptor signaling pathway	3	0.008901
BP term	GO:0042572	Retinol metabolic process	4	0.009418
CC term	GO:0005615	Extracellular space	63	9.65x10-17
CC term	GO:0031012	Extracellular matrix	28	2.46x10-14
CC term	GO:0070062	Extracellular exosome	87	1.68x10-12
CC term	GO:0005576	Extracellular region	61	4.86x10-12
CC term	GO:0005788	Endoplasmic reticulum lumen	20	4.73x10-11
CC term	GO:0005581	Collagen trimer	15	5.56x10-11
CC term	GO:0005604	Basement membrane	9	1.82x10-05
CC term	GO:0005578	Proteinaceous extracellular matrix	22	3.57x10-10
CC term	GO:0016324	Apical plasma membrane	16	2.29x10-05
CC term	GO:0009986	Cell surface	20	3.51x10-04
CC term	GO:0005887	Integral component of plasma membrane	34	0.004256
CC term	GO:0005886	Plasma membrane	79	0.004569
CC term	GO:0030141	Secretory granule	6	0.004319
CC term	GO:0031093	Platelet alpha granule lumen	5	0.008125
CC term	GO:0031090	Organelle membrane	6	0.008522
MF term	GO:0048407	Platelet-derived growth factor binding	6	2.55x10-07
MF term	GO:0005518	Collagen binding	8	2.37x10-05
MF term	GO:0050840	Extracellular matrix binding	6	3.05x10-05
MF term	GO:0005242	Inward rectifier potassium channel activity	4	0.002802
MF term	GO:0046332	SMAD binding	5	0.003328
MF term	GO:0005201	Extracellular matrix structural constituent	12	2.77x10-09
MF term	GO:0001758	Retinal dehydrogenase activity	3	0.004132
MF term	GO:0005178	Integrin binding	11	2.77x10-06
MF term	GO:0005509	Calcium ion binding	27	1.47x10-05
MF term	GO:0008201	Heparin binding	12	2.07x10-05
MF term	GO:0016491	Oxidoreductase activity	9	0.008547
MF term	GO:0008083	Growth factor activity	8	0.009105
KEGG pathway	hsa04512	ECM-receptor interaction	16	5.16x10-11
KEGG pathway	hsa04974	Protein digestion and absorption	14	7.73x10-09
KEGG pathway	hsa04510	Focal adhesion	18	2.67x10-07
KEGG pathway	hsa05146	Amoebiasis	10	1.63x10-04
KEGG pathway	hsa04971	Gastric acid secretion	8	4.23x10-04
KEGG pathway	hsa04151	PI3K-Akt signaling pathway	17	7.35x10-04
KEGG pathway	hsa00830	Retinol metabolism	7	0.00124
KEGG pathway	hsa00982	Drug metabolism-cytochrome P450	7	0.001703
KEGG pathway	hsa00980	Metabolism of xenobiotics by cytochrome P450	7	0.002628
KEGG pathway	hsa05204	Chemical carcinogenesis	7	0.003889

GO, Gene Ontology; KEGG, Kyoto Encyclopedia of Genes and Genomes; BP, biological processes; CC, cellular component; MF, molecular function.

Figure 2.

Gene Ontology terms and KEGG pathway enrichment analyses of 271 differentially expressed genes. Top 10 terms of enrichment for (A) BP, (B) CC and (C) MF. (D) Top 10 enriched KEGG pathways. KEGG, Kyoto Encyclopedia of Genes and Genomes; BP, biological process; CC, cellular component; MF, molecular function.

Based on the STRING prediction results, a PPI network with 211 nodes and 741 sides was constructed in Cytoscape (Fig. 3), and the number of segments connected to each gene in the figure represents its degree.

Figure 3.

Protein-protein interaction network of differentially expressed genes. Red indicates upregulated genes, and green represents downregulated genes.

Identification of six key genes

The two genes with the most nodes were FN1 and COL1A1. In the PPI network, FN1 was the most prominent, with the highest degree of connectivity at 52. The degree of connectivity of COL1A1 is 43 (Table IV). Expression of these two genes is upregulated in GC tissues. Additionally, of those DEGs shared among the four gene expression profiles, the two DEGs with the largest logFC and the two DEGs with the smallest logFC values were selected. The higher the logFC in the upregulated DEGs, the greater the increase in expression of the gene. Similarly, the lower the logFC values in the downregulated DEGs, the greater the decrease in expression of the gene. When sorting DEGs according to logFC, the logFC of GSE19826 was used as the standard, as chip GSE19826 represented a homogenous cancer tissue population at each Tumor-Node-Metastasis stage (25), which increases the accuracy of the expression profile (Table V). The two DEGs with the largest logFC values were INHBA (logFC=4.35) and CST1 (logFC=4.18) (Table VI). The two DEGs with the smallest logFC values were ATP4A (logFC=-6.46) and ATP4B (logFC=-5.91) (Table VII). Therefore, these six genes were selected as key genes.

Table IV

The 10 genes with the largest degree of connectivity in the protein-protein-interaction network.

Rank	Gene	Degree
1	FN1	52
2	COL1A1	43
3	COL1A2	38
4	COL3A1	37
5	FBN1	35
6	BGN	32
6	COL5A2	32
8	TIMP1	31
9	SPARC	30
10	THBS2	28

Table V

The expression data from GSE19826 in gastric cancer.

Tissue type	Accession no.	Title	Stage
Noncancer tissue	GSM495051	CB2008210-1N	n/a
Gastric cancer tissue	GSM495052	CB2008210-1T	II
Noncancer tissue	GSM495053	CB2008210-2N	n/a
Gastric cancer tissue	GSM495054	CB2008210-2T	IV
Noncancer tissue	GSM495055	CB2008210-3N	n/a
Gastric cancer tissue	GSM495056	CB2008210-3T	I
Noncancer tissue	GSM495057	CB2008210-4N	n/a
Gastric cancer tissue	GSM495058	CB2008210-4T	II
Noncancer tissue	GSM495059	CB2008210-5N	n/a
Gastric cancer tissue	GSM495060	CB2008210-5T	III
Noncancer tissue	GSM495061	CB2008210-6N	n/a
Gastric cancer tissue	GSM495062	CB2008210-6T	IV
Noncancer tissue	GSM495063	CB2008210-7N	n/a
Gastric cancer tissue	GSM495064	CB2008210-7T	IV
Noncancer tissue	GSM495065	CB2008210-9N	n/a
Gastric cancer tissue	GSM495066	CB2008210-9T	III
Noncancer tissue	GSM495067	CB2008210-12N	n/a
Gastric cancer tissue	GSM495068	CB2008210-12T	II
Noncancer tissue	GSM495069	CB2008210-13N	n/a
Gastric cancer tissue	GSM495070	CB2008210-13T	I
Noncancer tissue	GSM495071	CB2008210-14N	n/a
Gastric cancer tissue	GSM495072	CB2008210-14T	III
Noncancer tissue	GSM495073	CB2008210-15N	n/a
Gastric cancer tissue	GSM495074	CB2008210-15T	I
Normal gastric tissue	GSM495075	CB2008210-3C	n/a
Normal gastric tissue	GSM495076	CB2008210-5C	n/a
Normal gastric tissue	GSM495077	CB2008210-9C	n/a

Table VI

The 10 genes with the largest logFC values in GSE19826.

Rank	Name	LogFC
1	INHBA	4.35
2	CST1	4.18
3	COL11A1	4.11
4	FAP	3.91
5	COL10A1	3.72
6	FNDC1	3.27
6	COL8A1	3.17
8	SERPINH1	2.97
9	CDH3	2.95
10	THBS2	2.94

FC, fold change.

Table VII

The 10 genes with the smallest logFC values in GSE19826.

Rank	Name	LogFC
1	ATP4A	-6.46
2	ATP4B	-5.91
3	KCNE2	-5.88
4	AQP4	-5.81
5	GIF	-5.75
6	LIPF	-5.53
6	CHIA	-5.51
8	GKN1	-5.49
9	GKN2	-5.44
10	DPCR1	-4.83

FC, fold change.

Analysis of the six key genes in Oncomine

The Oncomine database was used to confirm the expression of the six key genes in 20 different types of cancer. The results showed that there were statistically significant differences in their expression. In the Oncomine database, there were no studies showing low expression of FN1, COL1A1, INHBA or CST1 in GC, but there were six, eight, seven and four studies showing increased expression, respectively. For ATP4A and ATP4B, the reverse was observed with no studies showing high expression, but seven and six studies, respectively, showing decreased expression (Fig. 4).

Figure 4.

mRNA expression of the six key genes in 20 different types of cancer. Cell color is determined by the best gene rank percentile for the analyses within the cell.

After comparing the expression levels of these six genes in cancerous and normal gastric tissue, the expression levels of FN1, COL1A1, INHBA and CST1 in GC tissues were significantly higher compared with the control group, and the expression levels of ATP4A and ATP4B in GC tissues were significantly lower compared with the control group (Table VIII; Fig. 5).

Table VIII

Additional information for the six key genes shown in Figure 5.

Author, year	Gene	Normal tissue samples	Gastric cancer samples	P-value	Fold Change	Published journal	(Refs.)
Chen et al, 2003	FN1	28	8	5.73x10-14	7.441	Molecular Biology of The Cell	(26)
Cui et al, 2011	COL1A1	80	80	1.81x10-15	3.201	Nucleic Acids Research	(28)
Cui et al, 2011	INHBA	80	80	5.17x10-13	3.043	Nucleic Acids Research	(28)
Cho et al, 2011	CST1	19	31	3.17x10-13	21.525	Clinical Cancer Research	(27)
Cho et al, 2011	ATP4A	19	20	4.73x10-17	-100.911	Clinical Cancer Research	(27)
D'Errico et al, 2009	ATP4B	31	26	6.15x10-19	-246.630	European Journal of Cancer	(11)

Figure 5.

Expression of six key genes in different gastric cancer gene chips in Oncomine. P<0.0001 and a |fold change|>2 were used as the threshold. Comparison of mRNA expression in cancerous vs. normal gastric tissue. (A) FN1, (B) COL1A1, (C) INHBA, (D) CST1, (E) ATP4A and (F) ATP4B.

In addition, meta-analyses of the six key genes in GC in the Oncomine database also supported the findings that expression of FN1, COL1A1, INHBA and CST1 is upregulated in GC, whereas expression of ATP4A and ATP4B is downregulated in GC (11,12,26-28). The studies and references involved are shown in Fig. 6. In the meta-analyses, P=-0.000, FC≥2.0 and gene rank ≤300 were selected as the cutoff criteria.

Figure 6.

Meta-analyses of the six key genes in gastric cancer in Oncomine. (A) FN1, (B) COL1A1, (C) INHBA, (D) CST1, (E) ATP4A and (F) ATP4B.

Survival analysis of the six key genes

To identify the prognostic value of the six potential key genes, overall survival curves based on differential expression of the six key genes were plotted using Kaplan-Meier plotter (Fig. 7). There were 1,440 patients with GC on the Kaplan-Meier plotter platform who were suitable for the analysis of overall survival. The curves indicate that overexpression of the six key genes is significantly associated with decreased overall survival times of patients with GC. However, it is worth noting that ATP4A and ATP4B were significantly downregulated in GC samples in the present study.

Figure 7.

Kaplan-Meier overall survival analyses of patients with gastric cancer based on expression of the six key genes. (A) FN1, (B) COL1A1, (C) INHBA, (D) CST1, (E) ATP4A, (F) ATP4B. HR, hazard ratio.

Discussion

GC is a complex heterogeneous disease with high incidence and mortality rates, and poses a serious threat to afflicted patients. Therefore, it is important to identify biomarkers that are meaningful for both diagnostic and prognostic assessment (29).

In the present study, 271 DEGs were screened, including 99 upregulated and 172 downregulated genes, by analyzing four gene expression profiles containing a combined 176 GC tissue samples and 82 normal gastric tissue samples. Of the causes of cancer-associated deaths, 90% are the result of metastasis (30). In the present study, GO enrichment results showed that the occurrence and development of GC was closely associated with metastasis. GO analysis indicated that DEGs were primarily associated with extracellular matrix organization, collagen catabolic process and cell adhesion. Collagen is the primary component of the extracellular matrix and of the interstitial microenvironment. Collagen can provide a scaffold for tumor cell growth and induce migration of tumor cells (31,32). There is evidence that collagen synthesis increases in the presence of a gastric tumor (33). Zhou et al (32) reported that collagen components are quantitatively and qualitatively reorganized in the tumor microenvironment of GC, and collagen width was identified as a useful prognostic indicator for GC (32). In addition, studies have shown that changes in cell-cell adhesion and cell-matrix adhesion can promote cancer cell metastasis (34). MF analysis showed that the DEGs were significantly enriched in platelet-derived growth factor binding. It has been demonstrated that inhibition of platelet-derived growth factor receptor-a can reduce the proliferation of gastrointestinal stromal tumor cells with mutant v-kit Hardy-Zuckerman 4 feline sarcoma viral oncogene homolog (KIT) by affecting the KIT-dependent transcription factor ETV1(35).

KEGG pathway analysis showed that the DEGs were primarily enriched in ECM-receptor interaction, protein digestion and absorption, and focal adhesion. ECM-receptor interaction serves a vital role in several types of cancer (36-38). The interaction between membrane receptors of tumor cells and ECM proteins serve an important role in tumor invasion and metastasis (39), and ECM-receptor interaction serve a crucial role in the process of tumor shedding, adhesion, degradation, movement and hyperplasia (38). In addition, the non-steroidal anti-inflammatory drug celecoxib may exhibit anti-GC effects by inhibiting the expression of various proteins and inhibiting leukocyte transendothelial migration and focal adhesion (40), which provides a possible mechanism for future investigations of the role of focal adhesion in GC and developing new anti-GC drugs.

The degree of connectivity of a gene in a PPI network reflects its association with GC. The greater the connectivity, the closer a gene is to the disease mechanism. The logFC values of DEGs reflects the level of up or downregulation of the gene. The higher the logFC values in the upregulated DEGs, the greater the degree of upregulation of the gene, and the lower the logFC values in the downregulated DEGs, the greater the degree of downregulation (41-43). Thus it was hypothesized that the DEGs with the highest and lowest logFC values would be the genes most closely associated with disease mechanisms.

In the present study, the two genes with the highest degree of connectivity in the PPI network, and the two DEGs with the largest and smallest logFC values, were all selected as key genes. These were FN1, COL1A1, INHBA, CST1, ATP4A and ATP4B. These six key genes were verified in the Oncomine database. Expression of FN1, COL1A1, INHBA and CST1 were upregulated in GC, and expression of ATP4A and ATP4B were downregulated, consistent with the results obtained from analysis of the GEO datasets. Furthermore, survival analysis showed that upregulation of the six key genes was significantly associated with worse overall survival, and downregulation of ATP4A and ATP4B expression predicted a more favorable prognosis for patients with GC, providing novel insights into potential GC treatment strategies.

FN1 was the gene with the highest degree of connectivity. It is expressed in a wide range of healthy plasmalemmas, lamina propria mucosae and smooth-muscle cell layers, and it is involved in a variety of cellular processes including embryogenesis, blood coagulation, wound healing, host defense and metastasis (44). As a glycoprotein involved in cell adhesion and migratory processes, FN1 is hypothesized to be associated with signaling pathways associated with cancer (13). Expression of FN1 is significantly increased in anti-chemotherapy osteosarcoma cell lines and tissues, and is associated with a poor prognosis (45). Knockdown of FN1 gene expression results in reduced cell proliferation, increased cellular senescence and apoptosis, and reduced migration and invasion, by blocking the activation of the PI3K/AKT signaling pathway (46). Furthermore, downregulation of FN1 inhibits proliferation, migration and invasion, and thus reduces progression of colorectal cancer (47). The results of the present study suggest that FN1 may be a potential biomarker and therapeutic target for diagnosis and treatment of GC, consistent with previous studies (13,48,49), and thus further confirming the significance of FN1 in GC.

COL1A1 is one of the most important components of the ECM, and it is highly expressed in most connective tissues and various human solid tumors (50). It is also the primary component of type I collagen, which serves a key role in tumor cell adhesion and invasion (51). A mechanistic study revealed that COL1A1 and COL1A2 affects angiogenesis in GC, and their expression is also significantly associated with progression of GC (52). In addition, Zhang et al (53) further confirmed that overexpression of COL1A1 promoted GC cell proliferation in vitro. These previous studies support the use of COL1A1 as a key potential GC biomarker in the present study.

INHBA is a member of the transforming growth factor-β (TGF-β) superfamily, which is closely associated with tumor proliferation and expression is upregulated in lung cancer (54), GC (12) and colon cancer (55), where INHBA expression is closely associated with their prognosis. In a study of GC, Chen et al (56) found that INHBA gene silencing reduced migration and invasion of GC cells by blocking the activation of the TGF-β signaling pathway. They suggested that INHBA was a potential target for GC therapy (56). Another study showed that INHBA mRNA expression in GC may be a useful prognostic biomarker for patients with stage II or III GC who receive adjuvant chemotherapy with S-1(57). The results of the present study support the conclusions drawn in these previous studies.

Cystatin SN (CST1) is a member of the type 2 cystatin superfamily, the primary role of which is to limit the proteolytic activity of cysteine proteases (58). The dysregulated expression of CST1 is hypothesized to be involved in several types of cancer, including cholangiocarcinoma (59), breast cancer (58), GC (60) and colorectal cancer (61). CST1 prevents cell aging and promotes cancer development by affecting the activity of cathepsin B (62). However, CST1 has not been analyzed using bioinformatics for survival prognosis in GC, to the best of our knowledge. Using multiple databases, the present study is the first to validate CST1 as a novel prognostic biomarker and a potential therapeutic target for treatment of GC.

ATP4A encodes the α subunit and ATP4B encodes the β subunit of the gastric H+, K+-ATPase, respectively. They regulate gastric acid secretion and, as a result, are targets for acid reduction (63). Fei et al (64) found that expression of ATP4A and ATP4B were significantly downregulated in patients with GC, but their expression was not significantly correlated with overall survival (64). In the present study, downregulation of ATP4A and ATP4B expression was associated with favorable overall survival in patients with GC. Downregulation of ATP4A and ATP4B mRNA expression in GC tissue is associated with the development of GC (65). Correa's Cascade is inversely associated with gastric acid secretion rate, the downregulation of ATP4A and ATP4B mRNA expression begins in the early stages of gastric mucosal lesions, and the expression of both is gradually decreased as Correa's cascade progresses (66). In addition, Helicobacter pylori (H. pylori) inhibits parietal acid secretion by downregulating the expression of ATP4A and ATP4B in gastric parietal cells prior to the formation of GC, suggesting that H. pylori is closely associated with the development of GC (67). Thus, it was hypothesized that ATP4A and ATP4B may inhibit the formation of GC. Survival analysis showed that ATP4A and ATP4B in GC are adverse prognostic factors for patients with GC, suggesting that upregulation is associated with progression of GC. However, studies have reported that the expression of ATP4A and ATP4B is not regulated by H. pylori in GC (68-70). Other studies have shown significant decreases in the abundance of Helicobacter and Neisseria, and significant increases in Achromobacter, Citrobacter, Phyllobacterium, Clostridium, Rhodococcus and Lactobacillus in gastric carcinoma in comparison with chronic gastritis (71,72). Additionally, the gastric microbiota composition in patients with gastric carcinoma is significantly different compared with patients with chronic gastritis (71). Therefore, it was hypothesized that the formation of an altered gastric microbiota composition may result in the expression of ATP4A and ATP4B to be passively upregulated as GC progresses. Further research is required to more accurately determine the biological function of ATP4A and ATP4B in GC.

Although several genes were identified as promising diagnostic and prognostic biomarkers for GC, the present study has the following limitations. First, the present study lacked experimental and clinical validation. Second, the possibility that different histological types may affect the accuracy of results cannot be eliminated. Thus, future bioinformatics analysis should be designed such that samples can be stratified by histological type. Finally, the sample size was relatively small for the RNA-Seq experiments, which may result in inaccuracies or results which are not completely representative of the wider populace. Therefore, it is necessary to use larger samples to perform bioinformatics analysis, and further experimental and clinical studies are required.

In conclusion, the present study used bioinformatics to analyze biological processes and signaling pathways closely associated with GC occurrence and development and identified FN1, COL1A1, INHBA and CST1 as promising diagnostic and prognostic biomarkers for GC patients. Additionally, the results of the survival analysis of ATP4A and ATP4B were inconsistent with other international studies. Therefore, further studies are required to assess the effects of ATP4A and ATP4B on GC initiation and development. Furthermore, experimental and clinical studies are required to validate the findings of the present study and determine the potential clinical value of these potential biomarkers.